litellm - 💡(How to fix) Fix RedisCluster drops socket timeout kwargs [2 pull requests]

Summary

RedisClusterCache does not pass socket_timeout or socket_connect_timeout through to redis.RedisCluster, so Redis cluster connections can hang even when those timeout values are configured in cache_params or router_settings.cache_kwargs.

This affects production Redis cluster deployments such as AWS ElastiCache Serverless / Valkey with cluster mode enabled.

Reproduction

Using the LiteLLM image with a TCP server that accepts connections on :6379 but never replies:

from litellm.caching.caching import RedisClusterCache

RedisClusterCache(
    startup_nodes=[{"host": "redis-blackhole", "port": 6379}],
    ssl=True,
    socket_timeout=1,
    socket_connect_timeout=2,
)

Expected: raises RedisClusterException quickly, consistent with direct redis.RedisCluster(...) behavior.

Observed: the LiteLLM RedisClusterCache construction continues hanging well past the configured timeout.

Direct redis-py behavior with the same kwargs does time out quickly:

redis.RedisCluster(... socket_timeout=1, socket_connect_timeout=2)
=> RedisClusterException: Redis Cluster cannot be connected. Please provide at least one reachable node: Timeout connecting to server
=> elapsed 1.01s

Root Cause

litellm._redis.init_redis_cluster() filters cluster kwargs through _get_redis_cluster_kwargs() before calling redis.RedisCluster(...). The allow-list includes values like ssl, username, and max_connections, but not socket_timeout or socket_connect_timeout, so those configured values are silently dropped for Redis cluster clients.

Proposed Fix

Add socket_timeout and socket_connect_timeout to _get_redis_cluster_kwargs() so they are forwarded to redis-py cluster clients.